3 ms·
Cerebras CEO: Our inference offering is 20x faster than Nvidia's
- JonChesterfield 2y agoAnyone know if you still have to program the thing in assembly using Cerebras' own engineers?
- billconan 2y ago> The tool costs 10c per million tokens for Lama 3.1 8B and 60c per million tokens for Llama 3 70B or roughly around 1/5 the price of GPU solutions. but how much to purchase the chip itself?